You are on page 1of 26

THE JOURNAL OF NAVIGATION (2011), 64, 495520.

f The Royal Institute of Navigation


doi:10.1017/S0373463311000099

Maritime Safety Standards and the


Seriousness of Shipping Accidents
Santiago Iglesias Baniela1 and Capt. Juan Vinagre R os2
1

(Universidad de La Coruna, Spain)


2
(Vessel Surveyor)
(E-mail : sbaniela@udc.es)

In our paper The Risk Homeostasis Theory1, it was accepted that the behaviour of people
involved in the operation of cargo carrying ships is conditioned to maximize the economic
benets of the amount of risk assumed in the transport. As a follow-up to that paper, the
objective of this one is to investigate the relationship between the level of compliance of the
cargo carrying vessels with international standards and the degree of severity of the incidents
they are involved in. For this purpose, we analyse the same sample of 2,584 cargo carrying
ships involved in incidents during 2005 and 2006 used in that investigation. The variables of
the Paris MoU to identify substandard ships are used again to measure the standard level of
ships and the degree of seriousness of incidents is determined by the number of days ships are
under repair.
KEY WORDS
1. Paris MoU.

2. Safety standards.

3. Shipping casualties.

1. I N T R O D U C T I O N. The globalization of the shipping industry makes it


dicult to establish safety incentive programmes which would lead to the reduction
of the frequency of shipping accidents2. Strict inspections, detentions in port, bans
and prohibitions to operate, etc. are common attempts, mainly adopted in the
USA, Paris and Tokyo MoU regions, to promote the operation of safer ships3. As a
result, the number of substandard vessels operating in those regions has decreased
during the last two decades. Nevertheless, since these policies do not eectively
reward ships or shipping companies with a lower ratio of accidents but simply
encourage them to comply with international standards, the rate of maritime mishaps goes beyond control. As can be observed in Figure 1, the total number
of shipping casualties varies, keeping pace with the global trade and the freight
markets4.
The accident rate per cargo carrying ship-year is thought to be a consequence of
a homeostatic control process in which the degree of caution of those engaged in
maritime transport determines the rate of accidents, and, as a consequence, this resulting rate determines the level of caution in the shipping operators behaviour5. In
this closed-loop process all types of accidents, near accidents and marine incidents
contribute to alter the perception of risk which those involved in maritime transport

S. I. B A N I E L A A N D C A P T. J. V. R I O S

496

V O L. 64

1200

Number of incidents

1000

800

600

400

200

0
1994 1995 1996 1997 1998 1999 2000 2001 2002 2003 2004 2005 2006 2007 2008 2009

Total Loss

Serious ex TL

Total

Figure 1. Total shipping casualties and total losses 19942009 (vessels>500 GT). Source:
www.iumi.com.

have ; but not all the mishaps have the same consequences for the economy, human
life or the environment.
Careful observation of shipping incidents leads us to perceive other aspects apart
from their mere occurrence. Historical review shows that dramatic maritime mishaps
are not as frequent nowadays as they used to be in the past6 ; in fact, if marine
accident statistics are analysed today, it will be seen that the incidence of total losses
of vessels is diminishing. In Figure 1 we observe that, while the tendency of the
frequency of all type of casualties has been moving upwards, total losses, observed
alone, have maintained a slightly downward trend during the same period. This fact
indicates that some factors have been operating to reduce the severity of incidents,
even though they have not altered the frequency of mishaps.
If we analyse the accidents that occurred to the ships of the world cargo carrying
eet during 2005 and 2006 and classify the mishaps using the necessary number of
days under repair as the level of seriousness of each incident, we obtain the results
shown in Figure 2, where we can see a known eect of the accident theory which
establishes that the frequency of incidents decreases as the damages caused increase7.
Very serious accidents, which caused repair for more than 300 days or total losses,
represent 4.6% of the total occurrences, while marine incidents, those in which ships
were 10 days or less under repair, represent 61.6%. This observation is consistent with
other industrial investigations8.
When we consider each accident type separately, the frequency changes. Depending
on how accidents such as collisions, groundings, res, contacts, damages to machinery,
etc. evolve, they may end up in simple incidents for some ships or serious accidents
for others. The risk homeostasis theory explains how human conduct alters the

N O. 3

MARITIME SAFETY STANDARDS

497

4.6 %

9.9 %

23.9 %

61.6 %

Type of accident
Marine incident [ days under repair 10 ]
Less serious [ 60 days under repair > 10 ]
Serious [ 300 days under repair > 60 ]
Very serious [ days under repair & total losses > 300 ]
Figure 2. Accidents by type occurred to the ships of the world cargo carrying eet in the period
20052006. Source: authors.

probability of occurrence of marine incidents ; but the dierences in their severity


prove the existence of factors that can alter the extent of the damages caused by the
behaviour of those engaged in the maritime transport.
2. M A R I T I M E S A F E T Y A N D P O L L U T I O N P R E V E N T I O N
STANDARDS AND THE LEVEL OF SERIOUSNESS OF
A C C I D E N T S. The experts of the maritime insurance industry have been expressing their concern about the increasing number of shipping casualties in recent
years9. Nevertheless, only some mishaps, mainly those causing damage to human life
or the environment, have been a matter of concern to the public and to sectors not
interested in maritime commercial activity. This concern, external to the maritime
business, perceptibly inuences the implementation of accident prevention rules.
It can be said that the mechanism that generates maritime safety and pollution prevention standards is not being moved by the occurrence of marine accidents but by
those at certain level of severity. Therefore, such standards should not be expected
to be directed to reducing the frequency of mishaps, but mainly their seriousness.
The presumption that the maritime safety and pollution prevention regulations are
mainly designed to prevent the most serious accidents is, in fact, reected in the rules
related to the obligation to investigate maritime casualties. The most important
conventions, associated with safety and pollution prevention, adopted by the

498

S. I. B A N I E L A A N D C A P T. J. V. R I O S

V O L. 64

International Maritime Organization (IMO) provide that each Administration must


conduct an investigation of any casualty occurring to any of its ships subject to the
provision of regulations and must supply the IMO with the ndings of such an investigation10. The Organization specically states it is more interested in those accidents
classied as very serious11 ; as a result, accident prevention policies are designed to
alleviate the consequences of this type of accident, such as the loss of human life or
environmental damage, which cannot be compensated by the shipping industry.
From a theoretical point of view, risk can be dened as a combination of the
likelihood of the occurrence of a hazardous event or exposure and the severity of the
damage caused12. Mathematically, it is usually dened as :
Risk=prD

(1)

Where p is the probability of occurrence of the mishap and D is the value of the
damage associated to the incident. Following this theoretical approach to the concept
of risk, two possible ways appear to reduce its quantity, either decreasing the severity
of the damage (D) or the probability of occurrence of incidents (p). Both are in fact
contained in the international maritime safety and pollution prevention regulations
to reduce the amount of risk in the activity of the world cargo carrying eet.
By going through the international standards, we can observe that many of them
are aimed at reducing the probability of occurrence of accidents. Rules related to the
prevention of collisions, information about ice, weather, derelicts and other perils to
navigation, the existence of ecient ship-borne navigational equipment and systems,
aids to navigation, automatisms, etc. are directed at reducing the probability of collisions, contacts or groundings. But the risk homeostasis theory considers that these
improvements in safety will be transformed in a more ecient and riskier activity due
to the behaviour of the people linked to the operation of ships, who do not nd any
economic motivation in being cautious. This natural impulse to optimize the efciency of maritime transport reduces the potential benets of safety improvements,
maintaining the probability of the occurrence of shipping accidents in the level of the
precedent stage.
Despite these unsuccessful actions to reduce the amount of risk by decreasing the
probability of occurrence of maritime accidents, there are many other standards which
are directed to reduce the severity of the damage caused by incidents. Regulations
concerned with subdivision and stability of ships13 do not alter the probability of
suering collisions, contacts or groundings, but they reduce the seriousness of these
events thus preventing, for example, the total loss of ships or severe pollution. The
existence of eective life-saving appliances and radio communication equipments
onboard and the good familiarity of the crew with such devices do not alter the
probability of the ship sinking, but these standards reduce the severity of these incidents thus preventing the loss of lives. Ecient re ghting appliances and re
resistant structures onboard do not reduce the probability of a re, but they are useful
to minimize the extent of the damage to the ship.
If we accept that the multiple combinations of the mandatory safety and pollution
prevention regulations hardly ever alter the probability of maritime accidents, but
that they are eective in reducing their severity, it should be expected that, during a
long period of observation of maritime mishaps, statistics would reect that substandard ships tend to suer more serious accidents than those of high standard,
although the tendency to suer an accident in both types of ships is similar due to the

N O. 3

MARITIME SAFETY STANDARDS

499

homeostatic process. The objective of this work is to analyse if this tendency is


satised.

3. A N A L Y S I S O F T H E A S S O C I A T I O N B E T W E E N T H E L E V E L
OF SAFETY AND POLLUTION PREVENTION STANDARDS OF
THE CARGO CARRYING SHIPS AND THE SERIOUSNESS OF
T H E I R A C C I D E N T S. To undertake the investigation of the relationship between the standard of vessels and the severity of mishaps, we have used the same
sample as in our previous paper14, which consists of 2,584 maritime incidents which
occurred during the years 2005 and 2006 worldwide and were reported by the
Lloyds Marine Intelligence Unit (LMIU). Only cargo carrying vessels of 100 GT
or above have been recorded. In order to know the level of seriousness of each incident, the number of days under repair has been recorded for each vessel with information obtained from LMIU. It is assumed that the higher the number of days is,
the more serious the accident will be. In case of total loss, 365 days have been recorded to facilitate calculations. The accidents have been put into four groups taking into consideration their severity, as shown in Figure 2.
The evaluation of the level of safety and pollution prevention standard of the
vessels involved in the accidents has been carried out following the criteria used by the
Paris MoU to identify substandard shipping. Thus, the ag, the classication society,
the age, the tonnage and the type of every ship involved have been recorded.
According to the Paris MoU Secretariat information, it is assumed that the prole of
the vessel of lowest safety and pollution prevention standard is that of a small-sized,
old general cargo or reefer ship, ying the ag of a country included in the black list
and either classied by a very low performance society or with unknown class.
Conversely, the highest standard ship is a big-sized, new and specialised vessel, ying
a white list ag and classied by a very high performance society.
Statistical analyses have been made with an appropriate computer program. Many
outputs, not essential to this investigation, have been omitted to reduce the extent of
this paper15.
3.1. Association between the ag of vessels and the seriousness of accidents. Following the information contained in the Paris MoU 2005 and 2006 annual
reports, all the ships involved in incidents during this period have been classied into
four groups according to the list of the ag they were ying at the time of the incident.
The black list has been subdivided into two groups: one for those countries included
in the very high risk group and the other for those in the high and medium risk one.
In Figure 3, the frequencies per type of accident for each ag group are shown. It
can be observed that there are perceptible dierences among the ag groups. The
more substandard the ag list is, the smaller the dierence of frequencies among
accident types will be. This observation leads us to assume that there is an association
between the level of safety and pollution prevention standard of the ships measured
by the performance of their ag and the level of severity of the incidents they are
involved in. To measure this relationship, we have made a table of contingency where
the frequency of accidents per cell is shown. The result is shown in Table 1, which
allows us to see the dierences among cells at a glance. The statistical signicance of
these dierences can be tested with the Pearsons chi-square test. If the proportion of
ships in the dierent columns varies between rows (and, therefore, vice versa) the

S. I. B A N I E L A A N D C A P T. J. V. R I O S

500

V O L. 64

Table 1. Contingency table showing the distribution of maritime accidents occurred during 2005 and 2006
per type of accident and Paris MoU ag group. Source: authors.
Contingency Table
Type of accident
Marine
incident
[f10 days]

Less serious
[60 odays
>10]

Serious
[300 odays
>60]

Very serious
[300 <days
& total losses]

Total

White
Grey
Black (High & Medium risk)
Black (Very High risk)

1010
486
69
28

379
171
45
22

113
101
31
11

34
46
20
18

1536
804
165
79

Total

1593

617

256

118

2584

Paris MoU ag list

Type of accident

1000

Marine incident
Less serious
Serious
Very serious

Count

800

600

400

200

0
White

Grey

High & Medium risk

Very High risk

Paris MoU flag list


Figure 3. Bar chart with the number of incidents by type and Paris MoU ag list of the cargo
carrying vessels involved in incidents during 2005 and 2006. Source: authors.

table shows association between the two variables. If there is no contingency, the two
variables are independent. In this case, a chi-square value of 156.972 with an associated signicance of less than 0.01 indicates that there is a relationship between the
ag of ships and the seriousness of the accidents they are involved in16.
To investigate this relationship, we use techniques of correspondence analysis17.
This method is traditionally applied to contingency tables and breaks up the
chi-square statistic associated to this table into orthogonal factors. The summary in
Table 2 shows a relationship between the row and column variables, and the number
of dimensions needed to display this relationship. The singular values are measures of

N O. 3

501

MARITIME SAFETY STANDARDS

Table 2. Summary of statistics and inertias of the correspondence analysis between the type of accident and
the Paris MoU ag group. Source: authors.

Dimension

Summary
Condence
Singular Value

Proportion of Inertia
*9 degrees of freedom

Correlation
Singular
Value
0.235
0.067
0.036

1
2
3

Inertia
0.055
0.004
0.001
0.061

Total

1.5

Chi Square

156.972

Sig.

Accounted
for

Cumulative

0.000*

0.906
0.073
0.021
1.000

0.906
0.979
1.000
1.000

Paris MoU flag list

Standard
Deviation
0.026
0.024

2
0.224

Type of accident
Very high risk

Dimension 2

1.0

Very serious

0.5
Less serious
White

0.0
Marine incident

High & Medium risk


Grey

- 0.5

- 1.0
- 1.0

Serious

- 0.5

0.0

0.5

1.0

1.5

2.0

Dimension 1
Figure 4. Bi-plot of the correspondence analysis between the type of accident and the Paris MoU
ag group of the ships involved. Source: authors.

association by dimension between the row and column variables, with larger values
indicating stronger relationships. The squares of the singular values equal the inertias,
which are capable of being added over dimensions. Here, the rst two dimensions
have 97.9% of the inertia in the solution, so the third dimension is unnecessary. The
graphical representations obtained with these statistical techniques have an easy interpretation : the shorter the distance (measured in chi-squared values) between points,
the higher their relationship.
The bi-plot representation of Figure 4 shows that ships registered in countries included in the white list by the Paris MoU tend to be involved in marine incidents and

502

S. I. B A N I E L A A N D C A P T. J. V. R I O S

V O L. 64

Table 3. Grouping of the Classication Societies of the investigated ships according to their level of
performance in the Paris MoU region. Source: Paris MoU.
Paris MoU Classication Society performance (20042006)
LEVEL
Very High
Det Norske
Veritas
Registro Italiano
Navale
Germanischer
Lloyd
China
Classication
Society
American Bureau
of Shipping

High
Nippon Kaiji
Kyokai
Lloyds Register

Medium

Low & unknown CS

Korean Register of
Shipping
Croatian register of
shipping
Polski Rejestr Statkow

Register of Shipping
(Korea DPR)
Register of shipping
(Albania)
INCLAMAR (Cyprus)

Turkish Lloyd

Hellenic Register of
Shipping

International Register of
Shipping (USA)

Russian Maritime
Register of
Shipping

Rinave

Unknown CS

Bureau Veritas

Indian register of
shipping
Bulgarski Koraben
Registar

less serious accidents. Vessels agged in grey list countries tend to suer serious
accidents and those in the black list (high and medium risk) are associated to serious
and very serious accidents. Finally, ships registered in countries included in the black
list (very high risk) tend to be clearly involved in very serious casualties.
3.2. The Classication Society (CS) of cargo carrying vessels and the severity of
their accidents. In this analysis, the level of safety and pollution prevention standard
of the ships involved in an accident will be evaluated according to the performance of
the society classifying the vessel. The degree of performance of societies when acting
as Recognized Organizations18 is assigned by the Paris MoU Secretariat. Following
this information, the Classication Societies of the investigated ships have been included in four groups as shown in Table 3. The information about the Classication
Society of the vessels at the time of the incident was taken from Lloyds Register
Fairplay and when no information appears, unknown class is reported.
By measuring the level of seriousness of accidents by the number of needed days
under repair and by grouping the ships involved in an accident by the level of performance of their Classication Society, we obtain the average days shown in
Figure 5. It can be observed that ships classied by low performance organizations or
with unknown class have an average of 90 days under repair. On the contrary, ships
classied by very high and high performance societies have a signicantly lower average (24 and 30 days respectively) than others. The dierences observed in the average (severity of accidents) depending on the level of performance of the
Classication Societies lead us to assume that there is an association between both
variables. In order to know if such dierences are statistically signicant, we have
used the non-parametric Kruskal-Wallis method for testing equality of population
medians among groups, without making any assumption about the distribution19.

N O. 3

503

MARITIME SAFETY STANDARDS

100

Days under repair

80

60

40

20

0
Very High
performance

High
performance

Medium
performance

Low
performance

CS performance in the Paris MoU region

Figure 5. Average days under repair for the cargo carrying vessels that suered any type of
incident during 2005 and 2006 and grouped by their Classication Society performance in the
Paris MoU region. Source: authors.

The mean ranks obtained are shown in Table 4, where we can appreciate that the
values of the groups of ships classied by high and very high performance societies
are very close. Nevertheless, the mean ranks in the group of ships classied by medium and low performance societies and ships with unknown class dier widely from
the other groups. The chi-square value obtained is 39.311 with an associated p-value
below 0.01. The output is shown in Table 5.
Signicant levels below 0.01 indicate that the group locations dier. This means
that ships classied by societies with high and very high performance in the Paris
MoU region tend to be involved in less serious accidents than those classied by low
performance ones or with unknown Classication Societies. To get a more detailed
examination of their relationship, we use categorical values for both variables obtaining the correspondence table of Table 6, where we can see the number of vessels
classied by type of accident and class society performance.
In Table 6, we have also displayed the output with the chi-squared and its signicant value, which conrms the association between variables. The inertias obtained
show that a two-dimension graph is suitable. The results obtained, seen in Figure 6,
show that ships classied by societies of high and very high performance are closely
associated to marine incidents and less serious accidents. Vessels classied in medium
performance societies tend to be involved in serious accidents and ships classied by
societies with low performance or with unknown class are associated either with
serious or very serious incidents. It can be observed that the higher the performance
of the society is, the less serious the accidents of the ships classied are.

S. I. B A N I E L A A N D C A P T. J. V. R I O S

504

V O L. 64

Table 4. Output of the Kruskal-Wallis test showing the N and mean rank values. Source: authors.
Ranks

Days under
repair

Classication Society performance

Mean Rank

Very high
High
Medium
Low & unknown CS

964
1061
193
366

Total

2584

1222.32
1270.47
1385.75
1492.03

Table 5. Output showing the chi-square statistic and the signicance. Source: authors.
Test Statistics [Kruskal Wallis Test ]
Grouping Variable: CS
Days under repair
39.311
3
0.000

Chi-Square
df
Asymp. Sig.

Table 6. Contingency table and summary statistics for the correspondence analysis of the relationship
between the classication societies of the ships involved in accidents during 2005 and 2006 and the type of
accident. Source: authors.
Correspondence Table
Type of accident

Classication Society performance


Very high
High
Medium
Low & unknown CS
Active Margin

Marine
incident

Less
serious

Serious

Very
serious

Active
Margin

629
672
108
184
1593

247
261
43
66
617

71
103
26
56
256

17
25
16
60
118

964
1061
193
366
2584

Dimension

Summary

1
2
3
Total

Proportion of Inertia

Condence Singular
Value
Correlation

Singular
Value
0.265
0.032
0.005

Inertia
0.070
0.001
0.000
0.071

Chi
Square

183.572

Sig.

Accounted
for

Cumulative

0.000

0.985
0.014
0.000
1.000

0.985
1.000
1.000
1.000

Standard
Deviation
0.025
0.019

2
0.098

N O. 3

505

MARITIME SAFETY STANDARDS

2.0
Type of accident

Dimension 2

Classification Society performance

1.0

Very serious

Very high
Marine incident

- 0.0

Low & unknown CS

Less serious
High

Medium
Serious

- 1.0
- 1.0

0.0

1.0

2.0

3.0

Dimension 1
Figure 6. Two-dimension graph showing the association between the Classication Society
performance in the Paris MoU region and the type of accident of the vessels involved in incidents
during 2005 and 2006. Source: authors.

3.3. The age of cargo carrying vessels and the severity of their accidents. With the
aim of investigating if there is any kind of relationship between the age of ships and
the seriousness of the accidents they are involved in, we have divided the cargo carrying vessels that suered accidents during 2005 and 2006 into four groups according
to their age at the time of the incident. As previously, the incidents were divided into
four groups according to their severity measured by the number of days under repair.
The result of this classication is shown in the cross-tabulation of Table 7, where the
columns represent the variable type of accident and the rows the variable age.
The dierences of values in the cells of the above table can be better appreciated in
the bar chart of Figure 7, where it can be observed that the number of very serious
casualties increases as the age of ships is greater. Taking this type of accident as a
percentage of the total accidents that occurred in the group, the dierences are more
signicant. This observation leads us to the intuitive idea that the newer the ships are,
the more triing are the accidents they are involved in.
To know if the dierences observed above are statistically signicant we have
made, as before, a chi-square test using the data of Table 7. With this statistic, the
hypothesis that the row and column variables in a cross-tabulation are independent
can be tested. The output of this method is shown in Table 8.
The low signicant values observed by the Pearson Chi-Square and the Likelihood
Ratio (below 0.01) indicate that there is a perceptible relationship between the two
variables although we do not know the strength or direction of the relationship. To
study these matters in depth, we use the correspondence analysis obtaining the inertias
shown in the output of Table 9. The value of the cumulative proportion of inertia for
two dimensions is 99.7%, which means that a bi-plot representation is adequate. This

S. I. B A N I E L A A N D C A P T. J. V. R I O S

506

V O L. 64

Table 7. Table showing the frequency of accidents occurred to the ships of the world cargo carrying eet
during 2005 and 2006 classied by type of accident and age of vessels. Source: authors.
Age of vessels Type of accident Crosstabulation
Count

Age of vessels

Type of accident

New vessel [f10 years ]


Middle-aged vessel
[>10 years f20 ]
Old vessel [>20 years f30 ]
Very old vessel [>30 years]
Total

Marine
incident
[o10 days]

Less serious
[60 odays
>10]

Serious
[300 odays
>60]

Very serious
[300 <days
& total losses]

Total

513
366

162
146

54
51

8
17

737
580

472
242

212
97

93
58

38
55

815
452

1593

617

256

118

2584

600

Type of accident

500

Marine incident
Less serious
Serious
Very serious

Count

400

300

200

100

0
New

Middle-aged

Old

Very old

Age of vessels
Figure 7. Bar chart with the number of incidents by type and age of the cross-tabulation of
Table 7. Source: authors.

is shown in Figure 8, where new vessels are closely associated to marine incidents.
Middle-aged ships tend to be involved in less serious accidents and marine incidents
and old vessels tend to suer less serious and serious accidents. Ships of more than
30 years old tend to be involved in a higher percentage of very serious casualties. The

N O. 3

507

MARITIME SAFETY STANDARDS


Table 8. Output with the statistical values and their signicance. Source: authors.
Chi-Square Tests
Value

df

110.222*
101.458
78.659
2584

Pearson Chi-Square
Likelihood Ratio
Linear-by-Linear Association
N of Valid Cases

Asymp. Sig. (2-sided)

9
0.000
9
0.000
1
0.000
*0 cells (0.0%) have expected count less than 5.
The minimum expected count is 20.64.

Table 9. Output with the statistics and inertias of the correspondence analysis of data in Table 7.
Source: authors.
Summary
Dimension

Proportion of Inertia

Condence Singular Value

*9 degrees of freedom
Correlation
Singular
Value
0.198
0.058
0.012

1
2
3
Total

Inertia
0.039
0.003
0.000
0.043

Chi
Square

110.222

Sig.

Accounted
for

Cumulative

0.000*

0.919
0.078
0.003
1.000

0.919
0.997
1.000
1.000

Age of vessels

Standard
Deviation

0.022
0.020

0.177

Type of accident

Dimension 2

1.0

0.5

Very serious
New ship

Very old ship

Marine incident
Middle-aged ship

0.0

Old ship
Serious
Less serious

- 0.5

- 1.0
- 1.0

- 0.5

0.0

0.5

1.0

1.5

2.0

Dimension 1
Figure 8. Bi-plot of the correspondence analysis between the type of accidents and the age of the
ships involved. Source: authors.

508

S. I. B A N I E L A A N D C A P T. J. V. R I O S

V O L. 64

Table 10. Vessels that suered incidents during 2005 and 2006 classied by size and type of accident.
Source: authors.
Size of ships Type of accident Cross-tabulation
Count

Size of ships

Type of accident

Total

Very big [>30 000 GT]


Big [>10 000 GT
f30 000]
Medium [>2000 GT
f10 000]
Small [f2000 GT]

Marine
incident
[f10 days]

Less serious
[60 odays
>10]

Serious
[300 odays
>60]

Very serious
[300 <days
& total losses]

Total

288
461

97
158

44
40

3
18

432
677

551

245

109

36

941

293

117

63

61

534

1593

617

256

118

2584

relationship between the age of ships and the level of seriousness of accidents is
statistically signicant and direct, which means that the older the vessel is, the more
serious the accident will be.
3.4. The size of cargo carrying vessels and the severity of their accidents. In order
to analyse the relationship between the size of ships and the seriousness of the accidents they are involved in, we have classied the vessels in four groups taking into
account their size measured by the tonnage GT and the severity of the accident as we
have done before. The result of the classication of ships that suered incidents
during 2005 and 2006 is shown in Table 10. If we represent the values of the cells of
the contingency table in a bar chart, we obtain Figure 9, where we can observe that the
bigger the ship is, the smaller the dierences among the type of accidents are.
The dierences observed in Figure 9 suggest a relationship between the type of
accident and the size of vessel variables. To know if this is statistically signicant, we
obtain the values of the Pearson Chi-Square and also the Likelihood Ratio for the
distribution of the accidents shown in the cross-tabulation of Table 10. These values
can be seen in the output of Table 11. The p-values associated to the obtained Pearson
Chi-Square and Likelihood Ratio statistics are less than 0.01. So we can conclude that
there is an association between the size of vessels and the seriousness of the accidents
they are involved in. We use the correspondence analysis, as we have done before with
the age of ships to understand the strength and direction of this relationship better.
The inertias obtained are shown in Table 12. With two dimensions we have 97.8% of
the cumulative proportion of inertia. Therefore, the bi-plot representation shown in
Figure 10 is adequate.
We can observe that bigger ships are associated with less serious accidents.
Conversely, small vessels are clearly associated with very serious casualties. The relationship between the size of ships and the severity of the accidents in which they are
involved is inverse : the bigger the ships are, the less serious the accidents will be.
3.5. The type of cargo carrying vessels and the severity of their accidents.
According to the Paris MoU information, and due to the circumstances of the

N O. 3

509

MARITIME SAFETY STANDARDS

600

Type of accident

500

Marine incident
Less serious
Serious
Very serious

Count

400

300

200

100

0
Very big

Big

Medium size

Small

Size of the vessel

Figure 9. Percentage of accidents of vessels that suered incidents during 2005 and 2006 classied
by size and type of accident. Source: authors.

shipping market, the less specialised vessels operating in the region (general cargo and
refrigerated cargo vessels) have a safety and pollution prevention standard level lower
than those types included in the specialised eet (tankers, bulk carriers, containers,
passenger ships, etc.). Assuming the signicant dierences in the grade of the standards
in the above mentioned groups, we have classied the vessels of the cargo carrying
eet by type in two groups. To investigate if the low standard types tend to be involved in more serious incidents than the high standard ones, we will use the number
of days under repair as an indication of the level of severity of the accidents. The
statistics obtained for both groups of vessels are shown in Table 13, which displays
the number of cases, mean value, standard deviation and standard error for the test
variable days under repair within categories dened by the grouping variable type of
ship. We can observe that the average number of days under repair obtained for nonspecialised ships is 53.88, while the average for specialised types is 29.09. This dierence is shown in Figure 11.
The values of the means of the two groups suggest that such a big dierence
cannot happen by chance and to know if this is statistically signicant, we use the
parametric independent samples t test procedure20 to compare the two group means
displayed in the group statistics table in Table 13. The output of this test is shown in
Table 14.
The value for the Levene test (less than 0.01) indicates that we must use the results
that do not assume equal variances for both groups of ships. The low value for the
t test (less than 0.01) indicates that there is a signicant dierence between the two

S. I. B A N I E L A A N D C A P T. J. V. R I O S

510

V O L. 64

Table 11. Output showing the statistics obtained for the distribution of the accidents shown in Table 9.
Source: authors.
Chi-Square Tests
Value

df

106.164*
100,397
57.474
2584

Pearson Chi-Square
Likelihood Ratio
Linear-by-Linear Association
N of Valid Cases

Asymp. Sig. (2-sided)

9
0.000
9
0.000
1
0.000
*0 cells (0.0%) have expected count less than 5.
The minimum expected count is 19.73.

Table 12. Output showing the inertias of the correspondence analysis of the contingency table in Table 10.
Source: authors.
Summary
Dimension

Proportion of Inertia

Condence Singular Value

*9 degrees of freedom
Correlation
Singular
Value
0.185
0.078
0.030

1
2
3
Total

Inertia
0.034
0.006
0.001
0.041

Chi
Square

106.164

Sig.

Accounted
for

Cumulative

0.000*

0.831
0.147
0.022
1.000

0.831
0.978
1.000
1.000

Standard
Deviation

0.022
0.018

0.049

1.5
Size of vessel

Type of accident

Dimension 2

1.0

0.5

Very serious

Big
Small

Marine incident
Very big

0.0

Less serious
Medium size

- 0.5

Serious

- 1.0
- 1.0

- 0.5

0.0

0.5

1.0

1.5

2.0

Dimension 1
Figure 10. Bi-plot of the correspondence analysis between the type of accidents and the size of the
ships involved. Source: authors.

N O. 3

511

MARITIME SAFETY STANDARDS

Table 13. Output with the statistics of the days under repair for specialised and non-specialised vessels of
the world cargo carrying eet that suered accidents during 2005 and 2006. Source: authors.
Group Statistics

Days under repair

Type of ship
Specialised eet
[Tankers, Bulk carriers,
Passenger ships,
Containers & RO-ROs......]
Non-specialised eet
[General cargo, Refrigerated cargo,
other cargo ships]

Number
of ships

Mean
[Days under
repair]

Std.
Deviation

Std. Error
Mean

1623

29.09

64.467

1.724

961

53.88

105.436

3.401

60

30

20

10

Tankers, Bulkcarriers,
Passenger ships, Containers
& ROROs ........

40

General cargo, Refrigerated cargo,


other cargo ships

Days under repair

50

0
Non-specialised fleet Specialised fleet
Type of ship

Figure 11. Average days under repair for specialised and non-specialised vessels of the world
cargo carrying eet that suered accidents during 2005 and 2006. Source: authors.

group means. The 99% condence interval for the mean does not contain zero ; this
also indicates that the dierence is signicant. To conrm this result, we have used the
non-parametric alternative to the t test : the Mann-Whitney U method. Table 15 lists
summary statistics. In this test, the two groups are combined and the data are ranked.
Mean rank lists the average of ranks for each group. If the groups have the same
location, these values should be similar. To test whether the locations dier, the
Mann-Whitney U Test or the Wilcoxon W Test are used and the results, which are

S. I. B A N I E L A A N D C A P T. J. V. R I O S

512

V O L. 64

Table 14. Output of the independent two sample t-test procedure of the days under repair for specialised
and non-specialised vessels of the world cargo carrying eet that suered accidents during 2005 and 2006.
Source: authors.
Independent Samples Test
Days under repair
Equal variances
assumed
Levenes Test for
Equality of Variances

F
Sig.

t-test for Equality


of Means

t
df
Sig. (2-tailed)
Mean Dierence
Std. Error Dierence
Lower
Upper

99% Condence
Interval of the
Dierence

137.371
0.000
x7.195
2582
0.000
x24.790
3.445
x33.671
x15.909

Equal variances
not assumed

x6.501
1459.844
0.000
x24.790
3.813
x34.625
x14.954

Table 15. Output of the statistics for the Mann-Whitney U test of the days under repair for specialised and
non-specialised vessels of the world cargo carrying eet that suered accidents during 2005 and 2006.
Source: authors.

Days under repair

Ranks

Type of ship

Mean
Rank

Sum of
Ranks

Specialised eet
[Tankers, Bulk carriers, Passenger ships,
Containers & RO-ROs......]

1623

1216.21

1 973 915.5

961

1421.34

1 365 904.5

Non-specialised eet
[General cargo, Refrigerated
cargo, other cargo ships]
Total

2584

shown in Table 16, yield identical conclusions. Small signicant values (in this case,
less than 0.01) indicate that the two groups have dierent locations.
The results obtained with either the parametric or the non-parametric tests lead us
to the same conclusion : the type of vessels pointed by the Paris MoU with a high level
of safety and pollution prevention standard tends to be involved in less serious accidents than those of low standard (reefers and general cargo ships). This result is
concordant with other investigations21.
4. A N A L Y S I S O F T H E C O M B I N E D E F F E C T S O F T H E F I V E
V A R I A B L E S. In Section 3 we have investigated the relationship between the
variables used by the Paris MoU to identify the level of safety and pollution prevention standard of the ships of the world cargo carrying eet and the level of

N O. 3

513

MARITIME SAFETY STANDARDS

Table 16. Output of the Mann-Whitney U and Wilkinson W tests of the days under repair for specialised
and non-specialised vessels of the world cargo carrying eet that suered accidents during 2005 and 2006.
Source: authors.
Test Statistics
[Grouping Variable: Type of ship]

Mann-Whitney U
Wilcoxon W
Z
Asymp. Sig. (2-tailed)

Days under repair


656 039.500
1 973 915.500
x6.813
0.000

Table 17. Ordinal values assigned to the variables. Source: authors.


Level

Level

Paris MoU ag list

White
Grey
Black (medium & low risk)
Black (high risk)

1
2
3
4

Age

New
Middle-aged
Old
Very old

1
2
3
4

Classication Society
performance

Very high
High
Medium
Low & Unknown CS

1
2
3
4

Size

Very big
Big
Medium
Small

1
2
3
4

Type

Specialised
Non-specialised

1
2

seriousness of the accidents they are involved in. These variables (ag, classication
society, age, size and type of ship) studied individually have shown a signicant association with the level of severity of accidents, pointing out that the lower the
standard of vessels is, the more serious the accidents are.
In order to know how the variables work together, we have analysed their combined eect. For this purpose, we give ordinal values to each of them taking into
account the direction of the level of safety and pollution prevention standard established in the Section 3 subparagraphs, as it is shown in Table 17.
If we assign the values to each variable for every cargo carrying ship which was
involved in an incident during 2005 and 2006 and we add the ve values, we obtain an
arbitrary scale with a score that goes from 5 to 18 for every vessel. This represents the
safety and pollution prevention standard level. It is assumed that increasing numbers
on the scale represent the increasing belief that the vessel belongs to the substandard
category and therefore the strength of conviction that it will be involved in a serious
accident ; decreasing numbers indicate the contrary. Thus, the higher the score of the
ship, the greater the likelihood of a serious accident. Considering a particular level of
seriousness of the accidents measured by the number of days under repair, we establish a binary classication, in which the number of days under or above an agreed
value labels the accident as not serious or serious respectively.
With the scores of the vessels and their binary classication, ROC curves are obtained and used to know how well the arbitrary level of the safety standard assigned
to each vessel predicts the actual type of accident occurred22. In this type of graph, the
area under the diagonal is 0.5 and if the curve obtained is close to this diagonal line, it

S. I. B A N I E L A A N D C A P T. J. V. R I O S

514

More than 60 days under repair

More than 10 days under repair


1.0

1.0

Positives:
374
Negatives: 2,210

Positives:
991
Negatives: 1,593
0.8

Sensitivity

0.8

Sensitivity

V O L. 64

0.6

0.4

0.6

0.4

0.2

0.2

Area under the curve = 0.675

Area under the curve = 0.588


0.0

0.0
0.0

0.2

0.4

0.6

0.8

0.0

1.0

0.2

0.4

0.6

0.8

1.0

1 - Specificity

1 - Specificity

More than 300 days under repair & total losses

More than 200 days under repair


1.0

1.0

Positives: 165
Negatives: 2,419
0.8

Sensitivity

Sensitivity

0.8

0.6

0.4

0.6

0.4

Positives: 118
Negatives: 2,466
0.2

0.2

Area under the curve = 0.785

Area under the curve = 0.812


0.0

0.0
0.0

0.2

0.4

0.6

1 - Specificity

0.8

1.0

0.0

0.2

0.4

0.6

0.8

1.0

1 - Specificity

Figure 12. ROC curves with the value of the area under the curve of the cargo carrying vessels
involved in incidents during 2005 and 2006 for four types of accidents in a binary classication
according to the number of days under repair. Source: authors.

means that serious or not serious accidents are randomly happening despite the score
of the vessels involved. The further the curve lies above the reference line, the more
ecient the prediction will be23. When the value of the area under the curve is close to
1, this means that near 100% of the substandard (high score) ships were involved in
serious accidents and also that near 100% of the high standard (low score) ones in not
serious incidents. In this analysis, the value of the area under the ROC curves is used
to give us an idea of the accuracy of the level of safety and pollution prevention
standard of ships (measured by the Paris MoU) to predict a particular degree of
severity of an accident.
In Figure 12 we have obtained the graphs for four dierent levels of seriousness
of accidents. In the rst ROC curve, we have classied the accidents in two groups

N O. 3

515

MARITIME SAFETY STANDARDS

ROC curve
1.0

Sensitivity

0.8

0.6

0.4

Areas under the curve


Flag = 0.696
CS = 0.744
Age = 0.717

0.2

Size = 0.707
Type = 0.647
0.0
0.0

0.2

0.4

0.6

0.8

1.0

1 - Specificity
Figure 13. ROC graph with the value of the area under the curves for each variable of the cargo
carrying vessels involved in very serious accidents during 2005 and 2006 (lost or with 300 days or
more under repair). Source: authors.

(not serious if the ship was 10 days or less under repair and serious if the repair took
more than 10 days). The number of ships involved in serious incidents was 991 and in
not serious 1,593. The area under the curve obtained is 0.588, which is close to the
diagonal (0.5). This means that accidents with this level of severity are randomly
happening to standard and substandard vessels. We can say that the mere occurrence
of incidents, without taking into account their severity, is happening to the ships by
chance despite their safety standard level.
When we increase the level of seriousness of incidents in a binary classication, the
areas under the ROC curves increase, showing that a higher percentage of substandard ships is involved in serious accidents and a higher percentage of standard
ones is involved in not serious incidents. If we agree that the level of severity of
incidents in total losses and accidents where the vessel involved was 300 days or
more under repair , the last graph in Figure 12 shows an area under the ROC curve
of 0.812, which is usually considered a signicant value. This indicates that the safety
and pollution prevention standard level assigned to the vessels, according to the Paris
MoU criteria, is useful to identify the prole of the ship prone to be involved in this
type of casualty, even though a certain level of uncertainty remains in the result.
Considering the level of seriousness in more than 300 days under repair and total
losses, we want to know the eect of each variable separately. For this purpose, we
obtain the ROC curves of Figure 13, where we observe that, taking into account the

516

S. I. B A N I E L A A N D C A P T. J. V. R I O S

V O L. 64

areas under the curves, the classication society and the age of the vessels are the most
inuential variables in the case of very serious casualties, followed by the size, the ag
and the type of ship.
The most appropriate way to investigate the contribution of each variable to explain this level of severity of accidents is using logistic regression models24. If we
consider the type of accident a categorical response variable with two outcomes, as
we have done with the ROC analysis, we can use the binary logistic regression procedure to determine whether the incident is more likely to be not serious (300 days
under repair or less) or serious (total loss or >300 days under repair). With this
analysis, we want to identify the variables which are indicative of ships prone to being
involved in a serious accident. We use the predictor categorical variables : the ag, the
class, the age, the size and the type of every ship with assigned values according to
Table 17. Then, we made a model to predict the probability of a given vessel being
involved in a serious accident (total loss or an incident of more than 300 days under
repair). If we call z the measure of the total contribution of every predictor variables,
the model can be dened as follows :
z=b0 +b1 Flag+b2 Class+b3 Age+b4 Size+b5 Type

(2)

Where b0, is the intercept and b1, b2, b3, b4, b5 are the regression coecients of the
variables. The value of each coecient is the size of contribution of their explanatory
variable. This model may describe the relationship between the above mentioned
variables and a binary response variable (type of accident), expressed as a probability
that has only two possible values : serious or not serious 25.
The output of Table 18 shows the coecients and associated statistics of the model
obtained with the computer when we introduce the 2,584 incidents labelled as serious or not serious. These coecients are estimated through an iterative maximum likelihood method. As can be seen, in the rst step the ve variables are in the
model. Taking into consideration the signicance of the Wald statistic, in the second
step the variable Size was removed, and in the third step, the variable Flag was
removed. The elimination of these two variables from the logit model does not mean
that they have no inuence on the type of accidents, but that their eect disappears
due to the interaction with the other variables. The actions of the Flag and the Size,
analysed alone, have shown a signicant relationship with the severity of the shipping
accidents26 in the last paragraphs.
The result obtained with the logistic regression analysis shows that the tendency of
a ship to be involved in a very serious accident is explained by the classication
society of the vessel, its age and type. The disappearance of the variable Flag seems to
reect what is happening nowadays in the maritime industry, where most of the
maritime Administrations entrust the inspection and survey of vessels to classication
societies of their condence; this practice links the Flag and the Class. The removal of
the variable Size from the model can also be explained by, on the one hand, the fact
that the bigger the ship, the stronger the pressure of the insurance industry to have the
vessel classied in a reliable society and, on the other hand, the tendency of bigger
ships to be specialised type vessels ; both facts link the variable Size to the variables
Class and Type. The situation of the shipping business makes the role of Class, Age
and Type more relevant in the logit model due to the interaction with the other
variables.

N O. 3

517

MARITIME SAFETY STANDARDS

Table 18. Output of SPSS showing estimated coecients and associated statistics of the logistic regression
model to explain accidents (more than 300 days under repair and total losses) occurring to cargo carrying
vessels during 2005 and 2006. Source: authors.

Step 1

Variables in the Equation


B

S.E.

Flag
Flag (1)
Flag (2)
Flag (3)

x0.856
x0.373
x0.478

0.376
0.345
0.375

Class
Class (1)
Class (2)
Class (3)

x1.381
x1.388
x0.520

0.343
0.295
0.315

Age
Age (1)
Age (2)
Age (3)

x1.321
x0.648
x0.425

0.415
0.314
0.244

Type (1)
Constant

x1.003
x0.118
x0.451
x0.732
x0.467

0.650
0.347
0.242
0.245
0.302

Flag
Flag (1)
Flag (2)
Flag (3)

x0.852
x0.380
x0.511

0.374
0.341
0.374

Class
Class (1)
Class (2)
Class (3)

x1.500
x1.502
x0.640

Step 2

df

Sig.

6.284
5.200
1.169
1.622
26.265
16.223
22.086
2.715

3
1
1
1

0.099
0.023
0.280
0.203
0.000
0.000
0.000
0.099

3
1
1
1

Exp(B)
0.425
0.689
0.620
0.251
0.250
0.595

11.682
10.116
4.267
3.036
5.347
2.380
0.116
3.482

3
1
1
1

0.009
0.001
0.039
0.081
0.148
0.123
0.734
0.062

8.907
2,386

1
1

0.003
0.122

3
1
1
1

0.325
0.271
0.306

6,233
5.189
1.237
1.867
37.337
21.238
30.617
4.378

0.101
0.023
0.266
0.172
0.000
0.000
0.000
0.036

0.223
0.223
0.527

Type (1)
Constant

x1.506
x0.729
x0.505
x0.792
x0.539

0.410
0.312
0.240
0.223
0.297

15.639
13.504
5.446
4.411
12.591
3.287

0.001
0.000
0.020
0.036
0.000
0.070

0.222
0.483
0.604
0.453
0.583

Class
Class (1)
Class (2)
Class (3)

x1.740
x1.608
x0.608

0.304
0.265
0.304

x1.632
x0.812
x0.533
x0.882
x0.938

0.404
0.306
0.240
0.208
0.182

Size
Size (1)
Size (2)
Size (3)

Age
Age (1)
Age (2)
Age (3)

Step 3

Wald

Age
Age (1)
Age (2)
Age (3)
Type (1)
Constant

3
1
1
1

3
1
1
1
3
1
1
1
1
1

50.774
32.725
36.855
4.004
19.196
16.298
7.026
4.943

3
1
1
1

0.000
0.000
0.000
0.045
0.000
0.000
0.008
0.026

18.055
26.514

1
1

0.000
0.000

3
1
1
1

0.267
0.523
0.654
0.367
0.889
0.637
0.481
0.627
0.427
0.684
0.600

0.175
0.200
0.544
0.196
0.444
0.587
0.414
0.392

518

S. I. B A N I E L A A N D C A P T. J. V. R I O S

V O L. 64

5. C O N C L U S I O N S. The empirical results obtained in this work lead us to


conclude that there is a signicant relationship between the safety standards and
the severity of shipping accidents. This association shows the tendency of substandard vessels to suer more serious accidents than others. The uncertainty, caused
by the fact that some very high standard ships suer very serious casualties whereas
some substandard vessels operate without being aected by a major accident, does
not contradict that previously mentioned relationship, but it shows that the capacity of standards to prevent the risks introduced by technological innovations in
the maritime transport is slowed by the economic rationale which, traditionally,
imposes the adoption of prevention measures after maritime accidents occur, not
before.
Analyses show that the higher the severity of the accidents is, the more accurate the
safety standards to predict them will be. This eciency diminishes until it disappears
in less serious incidents. As a consequence such standards are ineective to prevent
the occurrence of incidents, but ecient to mitigate their seriousness. The propensity
of a cargo carrying vessel to develop an incident into a more serious accident can be
explained by the degree of reliability of its classication society, its age and type.
For these analyses, the measure of the standard of ships was obtained with the
variables elaborated by the Paris MoU to identify substandard vessels. These factors
are based on the observation of the degree of compliance of vessels with international
safety and pollution prevention regulations. Thus, following the philosophy of the
ISO 9000 series and the International Safety Management Code27 and the results of
this paper, it can be assumed that, in a particular cargo carrying vessel, each level of
deviations from those regulations and procedures has an associated probability of
getting involved in a serious casualty after an incident occurs.
It is assumed that due to the homeostatic process, shipping accidents are randomly
happening to ships of the world cargo carrying eet, but the signicant relationship
between the level of the safety standard of vessels and the degree of seriousness of the
incidents they are involved in lead us to accept an accident causation theory, in which
maritime incidents start in a particular ship by chance and this initial event develops
into a nal stage, where the lower the standard of the ship is, the greater the extent of
the damage will be. This sequence follows an accident deviation model28.

REFERENCES
1
2

5
6

Baniela, S. I., Rios, J. V. (2010). The Risk Homeostasis Theory, The Journal of Navigation, 63, 607626.
Wilde, Gerald J. S. (1998). Safety Incentive Programmes. Encyclopaedia of Occupational Health and
Safety, 4th edition, Volume II, Part VIII, Chapter 60, ILO, Geneve [available from www.ilo.org/
safework_bookshelf/english].
The United States Coast Guard (Marine Safety Manual. Volume II: Materiel Inspection. Section D, Port
State Control); Memorandum of Understanding on Port State Control in the Asia-Pacic Region
(Section 3); Paris Memorandum on Port State Control (Section 3).
2009: Total Losses Down, Tonnage Lost Up. International Union of Marine Insurance (IUMI), press
release 29th March, 2010 [available from www.iumi.com].
Wilde, Gerald J. S. (1994). Target Risk. PDE Publications, Toronto.
Boisson, Philippe (1999). Safety at Sea: Policies, Regulations & International Law. Edition Bureau
Veritas, Paris, pp. 4555.
Skiba, Reinald (1998). Theoretical Principles of Job Safety. Encyclopaedia of Occupational Health
and Safety, 4th edition, Volume II, Part VIII, Chapter 56 ILO, Geneve [available from www.ilo.org/
safework_bookshelf/english].

N O. 3
8

MARITIME SAFETY STANDARDS

519

Heinrich, H. W. et al. (1959). Industrial Accident Prevention: A Scientic Approach. McGraw Hill Book
Company, New York.
9
Dramatic Increases in Merchant Ship Total and Partial Losses. International Union of Marine Insurance
(IUMI), press release 19th March, 2008 [available from www.iumi.com].
10
International Convention for the Prevention of Pollution from Ships, 1973 as modied by the Protocol
of 1978 (MARPOL 73/78), Article 12 and International Convention for the Safety of Life at Sea, 1974,
and its Protocol of 1978 (SOLAS 74/78), Chapter I, regulation 21.
11
See Casualty Investigation Code, Resolution MSC.255(84) (adopted on 16th May, 2008) where Chapter
6.1, related to very serious casualties, establishes A marine safety investigation shall be conducted into
every very serious marine casualty. Nevertheless, for other type of accidents, Chapter 17.1 establishes
A marine safety investigation should be conducted into marine casualties (other than very serious casualties which are addressed in chapter 6 of this Code) and marine incidents, by the ag State of a ship
involved, if it is considered [available from www.imo.org].
12
Jorgensen, Kirsten (1998). Concepts of Accident Analysis. Encyclopaedia of Occupational Health
and Safety (4th edition), Volume II, Part VIII, Chapter 56, ILO, Geneve [available from www.ilo.org/
safework_bookshelf/english].
13
These regulations are contained mainly in the SOLAS, MARPOL and Load Line Conventions.
14
Baniela, S. I., Rios, J. V. (2010). The Risk Homeostasis Theory, The Journal of Navigation, 63,
607626.
15
Statistical calculations have been made with the SPSS (Statistical Package for the Social Science)
Statistics program v.17. Many outputs of this program, mainly those concerning correspondence
analysis and ROC curves, are omitted to avoid an excessive extent of this paper.
16
Bulmer, M. G. (1979). Principles of Statistic, 2nd. Edition. Dover Publications, New York.
17
Benzecri, J. P., et al. (1992). Correspondence Analysis Handbook. Marcel Dekker, Inc., New York.
18
When a Classication Society is acting as a Recognised Organisation, it veries compliance with
national/international regulations adopted by a Flag State issuing the Statutory Certicates when appropriate and their responsibilities are dened in the Agreement with the Flag State unlike as when only
acting as a Classication Society verifying compliance with their own Rules, which is the only interpreter
of the requirements and it uses professional judgement to accept equivalent solution taking responsibility
and issuing the Class Certicates when appropriate.
19
Lecoutre, J. P., et al. (1987). Statistique non Parametrique et Robustesse. Economica, Paris.
20
Sheskin, David (2004). Handbook of Parametric and Nonparametric Statistical Procedures, 3rd. ed.
Chapman & Hall/CRC, Boca Raton (Florida).
21
Comparisons of general cargo ship losses and fatalities-Submitted by the Royal Institution of Naval
Architects (RINA). IMO MSC/77/25/4, 25th March, 2003 [available from www.imo.org].
22
A Receiver Operating Characteristic (ROC curve) is a graphical plot of the sensitivity vs. (1-specicity)
for a binary classier system as its discrimination threshold varies. A signal detection theory provides a
precise language and graphic notation for analysing decision making in the presence of uncertainty. By
tradition, the plot shows the false positive rate (1-specicity) on the X axis and the true positive rate
(sensitivity) on the Y axis. In this paper, sensitivity is the proportion of serious accidents with scores
greater than the cut-o and, 1-specicity is the proportion of not serious accidents in vessels with score
greater than the cut-o. Cut-o values go from 4 to 19 ; for instance, cut-o 4 is equivalent to assuming
that every vessel is involved in serious accidents and cut-o 18 is equivalent to assuming that every ship is
involved in not serious ones. Both extremes are unsatisfactory.
23
Egan, J. P. (1975). Signal Detection Theory and ROC Analysis. Academic Press, New York.
24
Kleinbaum, D. G., et al. (2002). Logistic Regression : A Self-Learning Text, 2nd. edition. SpringerVerlag, New York.
25
Statistical calculations to obtain b coecients were made assigning the values of the categorical variables
shown in Figure 28 to each ship, which means that for the variables Flag, Class, Age and Size the
coecients will have 3 values and for the variable Type only one value.
26
Model was estimated using a backward stepwise (Wald) method. In the output of Table 18, B is the
estimated coecient, with standard error S. E. The Wald statistic is signicant when less than 0.05, which
means that the parameter is useful to the model. Exp(B) is the predicted change in odds for a unit
increase in the predictor and when its value is less/greater than 1, it means that increasing values of the
variable correspond to decreasing/increasing odds of the events occurrence.
27
The International Safety Management Code (ISM) was made mandatory by Chapter IX of the SOLAS
Convention. It requires a safety management system to be established by the shipowner or any person

520

28

S. I. B A N I E L A A N D C A P T. J. V. R I O S

V O L. 64

who has assumed responsibility for the ship. The deviations from specied requirements are dened as
nonconformities.
Kjellen, Urban (1998). Accident Deviation Model. Encyclopaedia of Occupational Health and Safety,
4th edition, Volume II, Part VIII, Chapter 56, pp. 2224, ILO, Geneve [available from www.ilo.org/
safework_bookshelf/english].

You might also like